# https://github.com/haunt98/gill-cc3m
cp ../gill-cc3m/processed/cc3m_train.tsv datasets/cc3m_train.tsv
cp ../gill-cc3m/processed/cc3m_val.tsv datasets/cc3m_val.tsv
rm -rf data/cc3m
mkdir -p data/cc3m/training
mkdir -p data/cc3m/validation
cp ../gill-cc3m/output/training_01/*.jpg data/cc3m/training/
cp ../gill-cc3m/output/validation_01/*.jpg data/cc3m/validation/
Clone scripts/preprocess_sd_embeddings.py
with
sys.path.append('/path/to/gill')
python3 scripts/preprocess_sd_embeddings_local.py datasets/cc3m_train.tsv data/cc3m/training/clip_embs
python3 scripts/preprocess_sd_embeddings_local.py datasets/cc3m_val.tsv data/cc3m/validation/clip_embs
# TODO: Cache
python3 -u main.py \
--dataset=cc3m \
--val-dataset=cc3m \
--opt-version='facebook/opt-125m' \
--visual-model='openai/clip-vit-large-patch14' \
--exp-name='gill_exp' \
--log-base-dir='runs/' \
--batch-size=2 \
--val-batch-size=2 \
--precision='fp32' \
--print-freq=64 \
--epochs=2 \
--val_steps_per_epoch=2 \
--steps_per_epoch=2
python3 scripts/prune_model_ckpt.py runs/gill_exp